Download presentation
Presentation is loading. Please wait.
Published byJerome Benson Modified over 8 years ago
1
1 Shape Descriptors for Maximally Stable Extremal Regions Per-Erik Forss´en and David G. Lowe Department of Computer Science University of British Columbia Eleventh IEEE International Conference on Computer Vision ( ICCV – 2007 ) Reporter : Shih-Hao Wang
2
2 Outline Introduction Multi-Resolution Maximally Stable Extremal Regions ( MSER ) Experiment Model Experiment Conclusions
3
3 Introduction Affine-invariance Concept : originally features of images will not be changed after affine transformation Problem : When a scene is blurred or viewed from increasing distances, many details in the image disappear and different region boundaries are formed Use An affine invariant shape descriptor for maximally stable extremal regions (MSER) to reduce above- mentioned effect
4
4 Multi-Resolution Nested Vector Space : V 3 V 2 V 1 V 0 V 1 V 2 V 3
5
5 Multi-Resolution
6
6 Maximally Stable Extremal Regions Apply a series of thresholds – one for each grayscale level. Threshold the image at each level to create a series of black and white images. One extreme will be all white, the other all black. In between, blobs grow and merge.
7
7 Maximally Stable Extremal Regions Example
8
8 Multi-Resolution MSER Method : Step1 : Instead of detecting features (from step4) only in the input image Step2 : Construct a scale pyramid with one octave between scales Step3 : Detect MSERs separately at each resolution Step4 : Duplicate MSERs are removed by eliminating fine scale MSERs with similar locations and sizes as MSERs detected at the next coarser scale
9
9 Multi-Resolution MSER Scale pyramid The scale pyramid is constructed by blurring and resample with a Gaussian kernel, = 1.0 pixels.
10
10 Experiment (performance Measurement) Performance measure in this paper : inlier frequency k-th tentative correspondence is an inlier : inlier function is 1 Otherwise is 0
11
11 Experiment 3D scene correspondence evaluation The 56 accepted correspondences are shown in blue. Features from rejected correspondences are shown in green.
12
12 Experiment 3D scene results Performance of shape and texture descriptors on the scene
13
13 Experiment Planar and parallax-free scenes A scene with de-focus blur. The multi-resolution MSER provides better performance than using only the original resolution
14
14 Experiment Top image shows the 22 correspondences found using single resolution MSERs. Bottom image shows the 53 correspondences found using multi-resolution MSERs
15
15 Experiment Planar and parallax-free scenes A scene with scale change. Again, the multi-resolution MSER gives better performance
16
16 Experiment Top image shows the 22 correspondences found using single resolution MSERs. Bottom image shows the 33 correspondences found using multi-resolution MSERs
17
17 Conclusions New method provide better robustness to large scale changes and blurred images Improve matching performance over large scale changes and for blurred images
Similar presentations
© 2024 SlidePlayer.com Inc.
All rights reserved.